The 10 columns of `raw.Iris` fall into the four groups below. The six business columns come from the column profile I ran. The four `_`-prefixed lineage columns come from the dataset card (types only), because I did not profile them. **Identifier** - `Id` (BIGINT) is the primary key, with 0 nulls and values from 1 to 150. That matches one row per specimen across the 150 rows. **Categorical dimension** - `Species` (VARCHAR) is the only classifier, with 0 nulls and about 3 distinct values. It is the class label for the dataset. **Numeric metrics**, all DOUBLE, in centimetres, with 0 nulls: | Column | Min | Max | |---|---|---| | `SepalLengthCm` | 4.3 | 7.9 | | `SepalWidthCm` | 2.0 | 4.4 | | `PetalLengthCm` | 1.0 | 6.9 | | `PetalWidthCm` | 0.1 | 2.5 | The profiler labels these as `rough_kind = text`, but the physical type is DOUBLE and the semantic role is `measure`. I read that label as a profiling quirk. **Dates/times** - `_ingestion_timestamp` (TIMESTAMP) is the only temporal column. It records when each row was loaded, not when the flower was measured. The dataset has no observation date, so any time-based analysis would only describe load history. **Lineage metadata (not analytical)** - `_batch_id`, `_source_file` and `_source_system` are VARCHAR columns added at ingestion. They describe where and when the data was loaded. They are not identifiers of a specimen, not meaningful dimensions, and not metrics. In short, there is 1 identifier, 1 categorical dimension, 4 numeric metrics and 1 timestamp, plus 3 text lineage columns. For modelling, the four metrics are the features and `Species` is the target. `Id` and the lineage columns should be excluded.